pith. sign in

arxiv: 2606.05206 · v1 · pith:6JFT5LGDnew · submitted 2026-05-23 · 🧬 q-bio.NC · cs.AI· stat.AP

Ontology-constrained multi-LLM scoring of hypothesis support in the predictive processing literature

Hamed Nejat , Alexander Maier , Jesse Spencer-Smith , Andr\'e M. Bastos This is my paper

Pith reviewed 2026-06-30 11:47 UTC · model grok-4.3

classification 🧬 q-bio.NC cs.AIstat.AP
keywords predictive processingmulti-LLMontologyliterature synthesishypothesis scoringneuroscienceevidence mappingdisagreement analysis
0
0 comments X

The pith

Local multi-LLM councils score papers on predictive coding hypotheses using a fixed 36-concept glossary to create quantitative evidence maps.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper introduces a pipeline that has multiple local language models read individual studies from the predictive processing literature and score their support for three hypotheses based on a manually defined glossary of thirty-six concepts. The goal is to overcome the fragmentation in the field where studies span theory, electrophysiology, imaging, and behavior, making traditional meta-analysis difficult. The pipeline extracts evidence including from figure descriptions, assembles prompts, and validates against the glossary, then analyzes the resulting scores for agreements between studies and models. It also introduces a geometric measure of how spread out the scored studies are in hypothesis space. A reader would care if this provides a repeatable way to turn diverse papers into comparable numbers showing where evidence clusters or diverges.

Core claim

The central discovery is that a council of ten local language models, guided by an expert glossary of thirty-six concepts in the hypotheses of predictive suppression, feedforward error propagation, and ubiquity, can assign agreement scores to 31 studies across local and global oddball contexts. This yields auditable pairwise agreement data, cross-model comparisons, three-dimensional mappings of the hypothesis space, and a temperature metric for dispersion that is lower in local oddball contexts and higher in global ones, as well as vectors tracking changes between contexts.

What carries the argument

The multi-LLM scoring council constrained by the expert-defined 36-concept glossary, which serves as the common reference for evidence extraction and hypothesis scoring.

Load-bearing premise

The expert-defined 36-concept glossary is comprehensive and unbiased, and the ten local LLMs extract evidence and assign scores that faithfully reflect paper content without systematic model-specific distortions or misapplication of the ontology.

What would settle it

Running the same scoring task with human neuroscientists using the glossary on the 31 papers and observing that the LLM scores differ substantially and systematically from the human scores beyond inter-rater variability.

Figures

Figures reproduced from arXiv: 2606.05206 by Alexander Maier, Andr\'e M. Bastos, Hamed Nejat, Jesse Spencer-Smith.

Figure 1
Figure 1. Figure 1: A. Manual meta-studies (e.g., literature review, meta-analysis) are time-intensive and become harder to scale as the magnitude of the literature grows. B. The challenges with a traditional meta-analysis from the perspective of systems neuroscience. Heterogeneous data sources complicate unified statistical treatment. C. The LLM-assisted pipeline we propose here. AI-assisted literature review can ingest a la… view at source ↗
Figure 2
Figure 2. Figure 2: Benchmark overview. A 36-factor glossary defined the Hypothesis space (see Supplementary Tables S1-S3). Two separate contexts are defined, local and global oddballs. The corpus of papers (see Supplementary Table S5) were scored for agreement / disagreement with each hypothesis and in both contexts. Each agent performed these evaluations and saved the structured output in a table. Quantitative analysis of t… view at source ↗
Figure 3
Figure 3. Figure 3: The evaluation result of all papers (N=31) by all models (N=10). Scores are separated by context (LO = Local Oddball, GO = Global Oddball) and by Hypothesis (Supplementary Tables S1-S3, Hypothesis 1: Predictive Suppression; Hypothesis 2: Feedforward Error Propagation; Hypothesis 3: Ubiquity). Each dot is a separate score (total count = 1860). The mean and median of the scores are significantly positive (p … view at source ↗
Figure 4
Figure 4. Figure 4: Agent consistency measured as the mean squared distance between the aggregated scores of each of the 10 agents. Agent to agent distance is shown on the upper left. The mean in this comparison was 0.11. Mistral-nemo-12b was the model with the largest distance to the others. The mean squared distance of each model by context is shown for H1 (upper right), H2 (lower left), and H3 (lower right). The dotted lin… view at source ↗
Figure 5
Figure 5. Figure 5: Literature consistency (LC) visualization. The mean squared distance between every paper in the corpus is depicted. Warmer colors indicate lower MSD values or more agreement between studies. The highest values were observed for an outlier paper, Westerberg et al., 2025. 2.2. Direct Comparison of Local and Global Contexts between papers Further analysis of the distance scores between Westerberg, Xiong, et a… view at source ↗
Figure 6
Figure 6. Figure 6: Mean squared error values measuring the distance between each paper and the outlier paper (Westerberg, Xiong, et al., 2025), separately for the local oddball (upper subpanel) and global oddball (lower subpanel) contexts. Papers are ordered from left to right in order of increasing global-oddball MSD to Westerberg, Xiong, et al. (2025). 2.3. Three Dimensional Hypothesis Space Geometry Reveals a Constellatio… view at source ↗
Figure 7
Figure 7. Figure 7: Three-Dimensional view of the scores of each paper, with scores shown on the H1 (Predictive Suppression), H2 (Feedforward Prediction Errors), and H3 (Ubiquity) axes. The 3 lines emanating from each of the studies reflects the standard error of the mean of those scores across the 10 evaluator models. The local oddball context is shown on the left plot and the global oddball context is shown on the right. Mo… view at source ↗
Figure 8
Figure 8. Figure 8: Three-Dimensional vectors of each paper from the Local to the Global oddball position. Each paper’s vector is shown independently. The average movement (vector direction) without 12 [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
read the original abstract

Fragmentation is common in interdisciplinary fields with diverse methods and theoretical commitments. Predictive coding neuroscience is a clear example: its literature spans computational theory, electrophysiology, imaging, behavior, and modeling, creating a synthesis problem that conventional meta-analysis cannot easily resolve. Here, we describe a local multi-LLM pipeline for ontology-constrained literature synthesis. The pipeline reads papers, extracts evidence, incorporates figure descriptions, assembles constrained prompts, and validates outputs against an expert glossary. We manually defined a predictive-coding glossary of thirty-six concepts grouped into three hypotheses: predictive suppression, feedforward error propagation, and ubiquity. A council of ten local language models scored 31 studies according to their agreement or disagreement with each glossary factor across local and global oddball contexts. This enabled pairwise study-agreement analysis, cross-model comparison, and three-dimensional hypothesis-space mapping. Agreement was high for some hypotheses but weaker for others, revealing structured disagreement, particularly across local versus global oddball paradigms. We further define hypothesis-space temperature, a geometric dispersion metric measuring how compactly studies occupy the hypothesis space. Temperature was lower for local oddball contexts and higher for global oddball contexts, indicating greater dispersion in the latter. The scoring geometry also allowed us to estimate vectors of change between experimental contexts. These results demonstrate that local multi-LLM councils can produce auditable disagreement measurements that map heterogeneous literatures into quantitative evidence spaces. This framework may generalize to cross-study hypothesis mapping where conventional meta-analysis lacks a common comparison space.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript describes an ontology-constrained multi-LLM pipeline to synthesize the predictive processing literature. It manually defines a 36-concept glossary grouped into three hypotheses (predictive suppression, feedforward error propagation, and ubiquity), uses a council of ten local LLMs to score 31 papers on agreement with these concepts in local and global oddball contexts, performs pairwise agreement analysis, and introduces a 'hypothesis-space temperature' metric as a geometric dispersion measure. The results show structured disagreement, lower temperature in local contexts, and vectors of change between contexts, claiming this maps heterogeneous literatures into quantitative evidence spaces.

Significance. If the LLM scoring is shown to be reliable, this approach could offer a novel way to quantitatively map and compare studies in fragmented fields like predictive coding neuroscience, where conventional meta-analysis struggles due to diverse methods. The geometric temperature metric provides a new tool for assessing dispersion in hypothesis spaces, and the auditable nature of the multi-LLM council is a strength. However, without validation, its significance remains potential rather than demonstrated.

major comments (3)
  1. [Abstract] Abstract and Methods: the pipeline description reports no quantitative validation of LLM scoring accuracy against human raters (e.g., Cohen's kappa or equivalent inter-rater metrics) and supplies no error bars on the temperature metric. This is load-bearing because the pairwise agreement, hypothesis-space temperature, and context vectors all depend on the fidelity of the 36-concept scores.
  2. [Methods] Methods: no details are provided on how post-hoc decisions about context or concept grouping were made, and there is no ablation on model-specific biases in interpreting concepts such as predictive suppression or feedforward error propagation. This leaves open the possibility of systematic distortions in the derived evidence space.
  3. [Results] Results: the claim that temperature is lower for local oddball contexts and higher for global ones rests on the untested assumption that the ten LLMs faithfully extract evidence without model-specific distortions; the geometric definition alone does not address scoring validity.
minor comments (2)
  1. The three-dimensional hypothesis-space mapping would benefit from an explicit equation or diagram showing how the 36-concept scores are projected into the three hypothesis axes.
  2. Clarify whether the expert glossary was applied strictly during prompting or allowed any post-processing adjustments.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important aspects of validation and transparency. We address each major comment below, indicating planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract and Methods: the pipeline description reports no quantitative validation of LLM scoring accuracy against human raters (e.g., Cohen's kappa or equivalent inter-rater metrics) and supplies no error bars on the temperature metric. This is load-bearing because the pairwise agreement, hypothesis-space temperature, and context vectors all depend on the fidelity of the 36-concept scores.

    Authors: We agree that the absence of direct human validation is a limitation. In the revised manuscript we will add a dedicated limitations subsection discussing this gap and outlining future human-LLM comparison protocols. We will also compute and report bootstrap-derived confidence intervals for the temperature metric and include inter-model agreement statistics as an internal reliability proxy. revision: yes

  2. Referee: [Methods] Methods: no details are provided on how post-hoc decisions about context or concept grouping were made, and there is no ablation on model-specific biases in interpreting concepts such as predictive suppression or feedforward error propagation. This leaves open the possibility of systematic distortions in the derived evidence space.

    Authors: We will expand the Methods section with a new subsection detailing the rationale and process for context classification and concept grouping decisions. We will additionally conduct and report an ablation study that compares per-model score distributions and discusses potential biases for key concepts. revision: yes

  3. Referee: [Results] Results: the claim that temperature is lower for local oddball contexts and higher for global ones rests on the untested assumption that the ten LLMs faithfully extract evidence without model-specific distortions; the geometric definition alone does not address scoring validity.

    Authors: We accept that the temperature comparison should be qualified. In revision we will rephrase the relevant Results and Discussion passages to state that the metric captures dispersion within the LLM-derived scoring space and will cross-reference the new limitations and ablation sections. Full external validation remains outside the scope of the current study. revision: partial

Circularity Check

0 steps flagged

No significant circularity; new scoring pipeline with geometrically defined metrics

full rationale

The paper defines an expert glossary of 36 concepts, applies LLM scoring constrained by that glossary to 31 papers, then computes derived quantities (pairwise agreement, hypothesis-space temperature as geometric dispersion, context vectors) directly from the resulting coordinates. Temperature is introduced as a geometric definition applied to LLM-derived positions rather than fitted or self-referential. No equations or steps reduce by construction to prior fitted parameters, self-citations, or ansatzes from the authors' prior work. The central claim is the introduction of the pipeline itself, which is self-contained against external benchmarks and does not invoke load-bearing self-citations or uniqueness theorems. This is the normal non-circular outcome for a methodological contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The method rests on domain assumptions about LLM reliability and glossary validity plus one newly introduced geometric metric; no free parameters are fitted to data in the abstract description.

axioms (2)
  • domain assumption The manually defined 36-concept glossary accurately and exhaustively captures the three target hypotheses without significant omission or bias.
    The entire scoring and mapping procedure is built on this glossary.
  • domain assumption Local language models can extract relevant evidence from papers and assign consistent agreement scores according to the glossary without introducing systematic interpretive errors.
    This is the core mechanism that generates the quantitative evidence space.
invented entities (1)
  • hypothesis-space temperature no independent evidence
    purpose: Geometric dispersion metric that quantifies how compactly scored studies occupy the three-dimensional hypothesis space.
    Newly defined in the paper to summarize spread between local and global oddball contexts.

pith-pipeline@v0.9.1-grok · 5813 in / 1560 out tokens · 37512 ms · 2026-06-30T11:47:27.395703+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

87 extracted references · 21 canonical work pages · 3 internal anchors

  1. [1]

    (Verso, London, 2002)

    Feyerabend, P.Against Method: Outline of an Anarchistic Theory of Knowledge. (Verso, London, 2002). 20

    2002

  2. [2]

    Nature650, 857–863 (2026)

    Asai, A.et al.Synthesizing scientific literature with retrieval-augmented language models. Nature650, 857–863 (2026)

    2026

  3. [3]

    Intell.1, 14 (2025)

    Zhang, Y.et al.Exploring the role of large language models in the scientific method: from hypothesis to discovery.npj Artif. Intell.1, 14 (2025)

    2025

  4. [4]

    (2025) doi:10.48550/ARXIV.2502.18878

    Lu, Y.et al.Learning to Generate Structured Output with Schema Reinforcement Learning. (2025) doi:10.48550/ARXIV.2502.18878

    work page doi:10.48550/arxiv.2502.18878 2025

  5. [5]

    Nat Hum Behav9, 305–315 (2024)

    Luo, X.et al.Large language models surpass human experts in predicting neuroscience results. Nat Hum Behav9, 305–315 (2024)

    2024

  6. [6]

    Rao,R.P.N.&Ballard,D.H.Predictivecodinginthevisualcortex: afunctionalinterpretation of some extra-classical receptive-field effects.nature neuroscience2, 79–87 (1999)

    1999

  7. [7]

    V., Laughlin, S

    Srinivasan, M. V., Laughlin, S. B. & Dubs, A. Predictive coding: a fresh view of inhibition in the retina.Proc. R. Soc. Lond., B, Biol. Sci.216, 427–459 (1982)

    1982

  8. [8]

    H., Summerfield, C., Morin, E

    Bell, A. H., Summerfield, C., Morin, E. L., Malecek, N. J. & Ungerleider, L. G. Encoding of Stimulus Probability in Macaque Inferior Temporal Cortex.Current Biology26, 2280–2290 (2016)

    2016

  9. [9]

    B., Cadieu, C

    Issa, E. B., Cadieu, C. F. & DiCarlo, J. J. Neural dynamics at successive stages of the ventral visual stream are consistent with hierarchical error signals.Elife7, (2018)

    2018

  10. [10]

    The free-energy principle: a unified brain theory?Nature Reviews Neuroscience 11, 127–138 (2010)

    Friston, K. The free-energy principle: a unified brain theory?Nature Reviews Neuroscience 11, 127–138 (2010)

    2010

  11. [11]

    A., Shipp, S

    Adams, R. A., Shipp, S. & Friston, K. J. Predictions not commands: active inference in the motor system.Brain Struct Funct218, 611–643 (2013)

    2013

  12. [12]

    What Is Optimal about Motor Control?Neuron72, 488–498 (2011)

    Friston, K. What Is Optimal about Motor Control?Neuron72, 488–498 (2011)

    2011

  13. [13]

    T.et al.A Weighted and Directed Interareal Connectivity Matrix for Macaque Cerebral Cortex.Cereb

    Markov, N. T.et al.A Weighted and Directed Interareal Connectivity Matrix for Macaque Cerebral Cortex.Cereb. Cortexhttps://doi.org/10.1093/cercor/bhs270 (2012) doi:10.1093/cercor/bhs270

    work page doi:10.1093/cercor/bhs270 2012

  14. [14]

    M.et al.Canonical microcircuits for predictive coding.Neuron76, 695–711 (2012)

    Bastos, A. M.et al.Canonical microcircuits for predictive coding.Neuron76, 695–711 (2012)

    2012

  15. [15]

    Adams, R. A. Bayesian Inference, Predictive Coding, and Computational Models of Psychosis. inComputational Psychiatry175–195 (Elsevier, 2018). doi:10.1016/B978-0-12-809825- 7.00007-9

    work page doi:10.1016/b978-0-12-809825- 2018

  16. [16]

    Sterzer, P.et al.The Predictive Coding Account of Psychosis.Biological Psychiatry84, 634–643 (2018)

    2018

  17. [17]

    K., Suzuki, K

    Seth, A. K., Suzuki, K. & Critchley, H. D. An interoceptive predictive coding model of conscious presence.Front. Psychol3, 395 (2012). 21

    2012

  18. [18]

    & Litwin, P

    Miłkowski, M. & Litwin, P. Testable or bust: theoretical lessons for predictive processing. Synthese200, 462 (2022)

    2022

  19. [19]

    J., Nayak, A

    Bowman, H., Collins, D. J., Nayak, A. K. & Cruse, D. Is predictive coding falsifiable? Neuroscience & Biobehavioral Reviews154, 105404 (2023)

    2023

  20. [20]

    Behavioral and Brain Sciences36, 181–204 (2013)

    Clark,A.Whatevernext? Predictivebrains,situatedagents,andthefutureofcognitivescience. Behavioral and Brain Sciences36, 181–204 (2013)

    2013

  21. [21]

    (Oxford University Press, Oxford, 2013)

    Hohwy, J.The Predictive Mind. (Oxford University Press, Oxford, 2013)

    2013

  22. [22]

    Arnal, L. H. & Giraud, A.-L. Cortical oscillations and sensory predictions.Trends Cogn. Sci. (Regul. Ed.)16, 390–398 (2012)

    2012

  23. [23]

    Keller, G. B. & Mrsic-Flogel, T. D. Predictive Processing: A Canonical Cortical Computation. Neuron100, 424–435 (2018)

    2018

  24. [24]

    M., Lundqvist, M., Waite, A

    Bastos, A. M., Lundqvist, M., Waite, A. S., Kopell, N. & Miller, E. K. Layer and rhythm specificity for predictive routing.Proc Natl Acad Sci U S A117, 31459–31469 (2020)

    2020

  25. [25]

    C., Huang, Y

    Chao, Z. C., Huang, Y. T. & Wu, C.-T. A quantitative model reveals a frequency ordering of prediction and prediction-error signals in the human brain.Commun Biol5, 1–18 (2022)

    2022

  26. [26]

    C., Takaura, K., Wang, L., Fujii, N

    Chao, Z. C., Takaura, K., Wang, L., Fujii, N. & Dehaene, S. Large-Scale Cortical Networks for Hierarchical Prediction and Prediction Error in the Primate Brain.Neuron100, 1252-1266.e3 (2018)

    2018

  27. [27]

    Garrido,M.I.,Kilner,J.M.,Kiebel,S.J.&Friston,K.J.Evokedbrainresponsesaregenerated by feedback loops.Proc. Natl. Acad. Sci. U.S.A.104, 20961–20966 (2007)

    2007

  28. [28]

    Kok, P., Rahnev, D., Jehee, J. F. M., Lau, H. C. & de Lange, F. P. Attention Reverses the Effect of Prediction in Silencing Sensory Signals.Cerebral Cortex22, 2197–2206 (2011)

    2011

  29. [29]

    A.et al.Neural signature of the conscious processing of auditory regularities

    Bekinschtein, T. A.et al.Neural signature of the conscious processing of auditory regularities. Proceedings of the National Academy of Sciences106, 1672–1677 (2009)

    2009

  30. [30]

    J., van Mourik, T., Norris, D

    Kok, P., Bains, L. J., van Mourik, T., Norris, D. G. & de Lange, F. P. Selective Activation of the Deep Layers of the Human Primary Visual Cortex by Top-Down Feedback.Current Biology26, 371–376 (2016)

    2016

  31. [31]

    Kok, P., Jehee, J. F. M. & de Lange, F. P. Less is more: expectation sharpens representations in the primary visual cortex.Neuron75, 265–270 (2012)

    2012

  32. [32]

    R.et al.Predictions and errors are distinctly represented across V1 layers.Current Biology34, 2265-2271.e4 (2024)

    Thomas, E. R.et al.Predictions and errors are distinctly represented across V1 layers.Current Biology34, 2265-2271.e4 (2024)

    2024

  33. [33]

    Xiong,Y.(Sophy)etal.Propofol-mediatedlossofconsciousnessdisruptspredictiveroutingand local field phase modulation of neural activity.Proc. Natl. Acad. Sci. U.S.A.121, e2315160121 (2024). 22

    2024

  34. [34]

    A.et al.Hierarchical substrates of prediction in visual cortical spiking

    Westerberg, J. A.et al.Hierarchical substrates of prediction in visual cortical spiking. Preprint at https://doi.org/10.1101/2024.10.02.616378 (2024)

    work page doi:10.1101/2024.10.02.616378 2024

  35. [35]

    & Keller, G

    Zmarz, P. & Keller, G. B. Mismatch Receptive Fields in Mouse Visual Cortex.Neuron92, 766–772 (2016)

    2016

  36. [36]

    Bastos, G.et al.Top-down input modulates visual context processing through an interneuron- specific circuit.Cell Reports42, 113133 (2023)

    2023

  37. [37]

    P., Shymkiv, Y., Han, S., Yang, W

    Hamm, J. P., Shymkiv, Y., Han, S., Yang, W. & Yuste, R. Cortical ensembles selective for context.Proc Natl Acad Sci U S A118, e2026179118 (2021)

    2021

  38. [38]

    Nature645, 192–200 (2025)

    Findling, C.et al.Brain-wide representations of prior information in mouse decision-making. Nature645, 192–200 (2025)

    2025

  39. [39]

    Rodriguez,N.Y.,Ahuja,A.,Basu,D.,McKim,T.H.&Desrochers,T.M.DifferentSubregions of Monkey Lateral Prefrontal Cortex Respond to Abstract Sequences and Their Components. J. Neurosci.44, (2024)

    2024

  40. [40]

    Preprint at https://doi.org/10.48550/arXiv.2504.09614 (2026)

    Aizenbud, I.et al.Neural mechanisms of predictive processing: a col- laborative community experiment through the OpenScope program. Preprint at https://doi.org/10.48550/arXiv.2504.09614 (2026)

    work page doi:10.48550/arxiv.2504.09614 2026

  41. [41]

    in (Montreal, Canada, 2024)

    MulveyA.G.etal.DeficitsinGABAergicInhibitoryInterneuronsinSchizophreniaareSpecific to Cell type, Cortical layer, and Brain Region. in (Montreal, Canada, 2024)

    2024

  42. [42]

    2025.06.13.25329541 Preprint at https://doi.org/10.1101/2025.06.13.25329541 (2026)

    Cao, C.et al.Automation of Systematic Reviews with Large Language Models. 2025.06.13.25329541 Preprint at https://doi.org/10.1101/2025.06.13.25329541 (2026)

    work page doi:10.1101/2025.06.13.25329541 2025

  43. [43]

    & Susnjak, T

    Li, L., Mathrani, A. & Susnjak, T. Transforming Evidence Synthesis: A Systematic Review of the Evolution of Automated Meta-Analysis in the Age of AI. Preprint at https://doi.org/10.48550/arXiv.2504.20113 (2025)

    work page doi:10.48550/arxiv.2504.20113 2025

  44. [44]

    Song, Z.et al.Evaluating Large Language Models in Scientific Discovery.arXiv.org https://arxiv.org/abs/2512.15567v1 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  45. [45]

    Lu, C.et al.Towards end-to-end automation of AI research.Nature651, 914–919 (2026)

    2026

  46. [46]

    Li, J., Zhang, Q., Yu, Y., Fu, Q. & Ye, D. More Agents Is All You Need. Preprint at https://doi.org/10.48550/arXiv.2402.05120 (2024)

    work page doi:10.48550/arxiv.2402.05120 2024

  47. [47]

    & Shah, N

    Luo, Z., Kasirzadeh, A. & Shah, N. B. The More You Automate, the Less You See: Hidden Pitfalls of AI Scientist Systems. Preprint at https://doi.org/10.48550/arXiv.2509.08713 (2025)

    work page doi:10.48550/arxiv.2509.08713 2025

  48. [48]

    Aygün, E.et al.An AI system to help scientists write expert-level empirical software.Nature https://doi.org/10.1038/s41586-026-10658-6 (2026) doi:10.1038/s41586-026-10658-6

    work page doi:10.1038/s41586-026-10658-6 2026

  49. [49]

    M., Xiong, Y

    Gabhart, K. M., Xiong, Y. (Sophy) & Bastos, A. M. Predictive coding: a more cognitive process than we thought?Trends in Cognitive Sciences0, (2025). 23

    2025

  50. [50]

    S., Tang, H., Sussman, E

    Solomon, S. S., Tang, H., Sussman, E. & Kohn, A. Limited Evidence for Sensory Prediction Error Responses in Visual Cortex of Macaques and Humans.Cerebral Cortex31, 3136–3152 (2021)

    2021

  51. [51]

    V.et al.Auditory Predictive Coding across Awareness States under Anesthesia: An Intracranial Electrophysiology Study.J

    Nourski, K. V.et al.Auditory Predictive Coding across Awareness States under Anesthesia: An Intracranial Electrophysiology Study.J. Neurosci.38, 8441–8452 (2018)

    2018

  52. [52]

    Wacongne, C.et al.Evidence for a hierarchy of predictions and prediction errors in human cortex.Proceedings of the National Academy of Sciences108, 20754–20759 (2011)

    2011

  53. [53]

    Biases in the Blind Spot: Detecting What LLMs Fail to Mention

    Arcuschin, I., Chanin, D., Garriga-Alonso, A. & Camburu, O.-M. Biases in the Blind Spot: Detecting What LLMs Fail to Mention.arXiv.orghttps://arxiv.org/abs/2602.10117v4 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  54. [54]

    & Sethi, N

    Beatty, D., Masanthia, K., Kaphol, T. & Sethi, N. Revealing Hidden Bias in AI: Lessons from Large Language Models.arXiv.orghttps://arxiv.org/abs/2410.16927v1 (2024)

    work page arXiv 2024

  55. [55]

    karpathy/llm-council

    Andrej. karpathy/llm-council. (2026)

    2026

  56. [56]

    Yang, H.et al.Large Language Model Synergy for Ensemble Learning in Medical Question Answering: Design and Evaluation Study.Journal of Medical Internet Research27, e70080 (2025)

    2025

  57. [57]

    & Luo, P

    Li, Z., Tian, H., Luo, L., Cao, Y. & Luo, P. DeepRead: Document Structure-Aware Reasoning to Enhance Agentic Search. Preprint at https://doi.org/10.48550/arXiv.2602.05014 (2026)

    work page doi:10.48550/arxiv.2602.05014 2026

  58. [58]

    A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT

    White, J.et al.A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT. Preprint at https://doi.org/10.48550/arXiv.2302.11382 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2302.11382 2023

  59. [59]

    Rao,R.P.N.&Ballard,D.H.Predictivecodinginthevisualcortex: afunctionalinterpretation of some extra-classical receptive-field effects.Nat Neurosci2, 79–87 (1999)

    1999

  60. [60]

    & Keller, G

    Attinger, A., Wang, B. & Keller, G. B. Visuomotor Coupling Shapes the Functional Develop- ment of Mouse Visual Cortex.Cell169, 1291-1302.e14 (2017)

    2017

  61. [61]

    & Richards, B

    Bakhtiari, S., Mineault, P., Lillicrap, T., Pack, C. & Richards, B. The functional specialization of visual cortex emerges from training parallel pathways with self-supervised predictive learning. inAdvances in Neural Information Processing Systemsvol. 34 25164–25178 (Curran Associates, Inc., 2021)

    2021

  62. [62]

    A.et al.Neural signature of the conscious processing of auditory regularities

    Bekinschtein, T. A.et al.Neural signature of the conscious processing of auditory regularities. Proc. Natl. Acad. Sci. U.S.A.106, 1672–1677 (2009)

    2009

  63. [63]

    D., Aldea, A

    Furutachi, S., Franklin, A. D., Aldea, A. M., Mrsic-Flogel, T. D. & Hofer, S. B. Cooperative thalamocortical circuit mechanism for sensory prediction errors.Nature633, 398–406 (2024)

    2024

  64. [64]

    Garrett, M.et al.Experience shapes activity dynamics and stimulus coding of VIP inhibitory cells.eLife9, e50340 (2020). 24

    2020

  65. [65]

    W., Pemberton, J., Mellor, J

    Greedy, W., Zhu, H. W., Pemberton, J., Mellor, J. & Ponte Costa, R. Single-phase deep learning in cortico-cortical networks.Advances in Neural Information Processing Systems35, 24213–24225 (2022)

    2022

  66. [66]

    eLife9, e57541 (2020)

    Hertäg,L.&Sprekeler,H.Learningpredictionerrorneuronsinacanonicalinterneuroncircuit. eLife9, e57541 (2020)

    2020

  67. [67]

    Jiang, L. P. & Rao, R. P. Dynamic predictive coding: A model of hierarchical sequence learning and prediction in the neocortex.PLOS Computational Biology20, e1011801 (2024)

    2024

  68. [68]

    B., Bonhoeffer, T

    Keller, G. B., Bonhoeffer, T. & Hübener, M. Sensorimotor mismatch signals in primary visual cortex of the behaving mouse.Neuron74, 809–815 (2012)

    2012

  69. [69]

    J., Daunizeau, J

    Kiebel, S. J., Daunizeau, J. & Friston, K. J. A Hierarchy of Time-Scales and the Brain.PLoS Comput Biol4, e1000209 (2008)

    2008

  70. [70]

    B. Lao-Rodríguez, A.et al.Neuronal responses to omitted tones in the auditory brain: A neu- ronal correlate for predictive coding.Science Advanceshttps://doi.org/10.1126/sciadv.abq8657 (2023) doi:10.1126/sciadv.abq8657

    work page doi:10.1126/sciadv.abq8657 2023

  71. [71]

    Lee, K., Pennartz, C. M. A. & Mejias, J. F. Cortical networks with multiple interneuron types generate oscillatory patterns during predictive coding.PLOS Computational Biology21, e1013469 (2025)

    2025

  72. [72]

    A., Rudelt, L., Wibral, M

    Mikulasch, F. A., Rudelt, L., Wibral, M. & Priesemann, V. Where is the error? Hierarchical predictive coding through dendritic error computation.Trends in Neurosciences46, 45–59 (2023)

    2023

  73. [73]

    K., Anastasiades, P., Hertäg, L

    Nejad, K. K., Anastasiades, P., Hertäg, L. & Costa, R. P. Self-supervised predictive learning accounts for cortical layer-specificity.Nature Communications16, 6178 (2025)

    2025

  74. [74]

    Payeur, A., Guerguiev, J., Zenke, F., Richards, B. A. & Naud, R. Burst-dependent synaptic plasticity can coordinate learning in hierarchical circuits.Nat Neurosci24, 1010–1019 (2021)

    2021

  75. [75]

    Rao, R. P. N. A sensory–motor theory of the neocortex.Nat Neurosci27, 1221–1235 (2024)

    2024

  76. [76]

    & Senn, W

    Sacramento, J., Ponte Costa, R., Bengio, Y. & Senn, W. Dendritic cortical microcircuits approximate the backpropagation algorithm. inAdvances in Neural Information Processing Systemsvol. 31 (Curran Associates, Inc., 2018)

    2018

  77. [77]

    Spratling, M. W. Reconciling predictive coding and biased competition models of cortical function.Front. Comput. Neurosci2, 4, (2008)

    2008

  78. [78]

    Spratling, M. W. Predictive coding as a model of response properties in cortical area V1.J. Neurosci. Off. J. Soc. Neurosci30, 3531–3543 (2010)

    2010

  79. [79]

    B.et al.A Role for Somatostatin-Positive Interneurons in Neuro-Oscillatory and Information Processing Deficits in Schizophrenia.Schizophrenia Bulletin47, 1385–1398 (2021)

    Van Derveer, A. B.et al.A Role for Somatostatin-Positive Interneurons in Neuro-Oscillatory and Information Processing Deficits in Schizophrenia.Schizophrenia Bulletin47, 1385–1398 (2021). 25

    2021

  80. [80]

    & Dehaene, S

    Wacongne, C., Changeux, J.-P. & Dehaene, S. A Neuronal Model of Predictive Coding Accounting for the Mismatch Negativity.J. Neurosci.32, 3665–3678 (2012)

    2012

Showing first 80 references.